1 /*
2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "compiler/compilerDefinitions.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/collectedHeap.hpp"
32 #include "gc/shared/tlab_globals.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/markWord.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/os.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40
41 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
42 const int aligned_mask = BytesPerWord -1;
43 const int hdr_offset = oopDesc::mark_offset_in_bytes();
44 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
45 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
46 Label done;
47 int null_check_offset = -1;
48
49 verify_oop(obj);
50
51 // save object being locked into the BasicObjectLock
52 movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
53
54 null_check_offset = offset();
55
56 if (DiagnoseSyncOnValueBasedClasses != 0) {
57 load_klass(hdr, obj, rscratch1);
58 movl(hdr, Address(hdr, Klass::access_flags_offset()));
59 testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
60 jcc(Assembler::notZero, slow_case);
61 }
62
63 // Load object header
64 movptr(hdr, Address(obj, hdr_offset));
65 // and mark it as unlocked
66 orptr(hdr, markWord::unlocked_value);
67 // save unlocked object header into the displaced header location on the stack
68 movptr(Address(disp_hdr, 0), hdr);
69 // test if object header is still the same (i.e. unlocked), and if so, store the
70 // displaced header address in the object header - if it is not the same, get the
71 // object header instead
72 MacroAssembler::lock(); // must be immediately before cmpxchg!
73 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
74 // if the object header was the same, we're done
75 jcc(Assembler::equal, done);
76 // if the object header was not the same, it is now in the hdr register
77 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
78 //
79 // 1) (hdr & aligned_mask) == 0
80 // 2) rsp <= hdr
81 // 3) hdr <= rsp + page_size
82 //
83 // these 3 tests can be done by evaluating the following expression:
84 //
85 // (hdr - rsp) & (aligned_mask - page_size)
86 //
87 // assuming both the stack pointer and page_size have their least
88 // significant 2 bits cleared and page_size is a power of 2
89 subptr(hdr, rsp);
90 andptr(hdr, aligned_mask - (int)os::vm_page_size());
91 // for recursive locking, the result is zero => save it in the displaced header
92 // location (NULL in the displaced hdr location indicates recursive locking)
93 movptr(Address(disp_hdr, 0), hdr);
94 // otherwise we don't care about the result and handle locking via runtime call
95 jcc(Assembler::notZero, slow_case);
96 // done
97 bind(done);
98
99 inc_held_monitor_count();
100
101 return null_check_offset;
102 }
103
104 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
105 const int aligned_mask = BytesPerWord -1;
106 const int hdr_offset = oopDesc::mark_offset_in_bytes();
107 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
108 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
109 Label done;
110
111 // load displaced header
112 movptr(hdr, Address(disp_hdr, 0));
113 // if the loaded hdr is NULL we had recursive locking
114 testptr(hdr, hdr);
115 // if we had recursive locking, we are done
116 jcc(Assembler::zero, done);
117 // load object
118 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
119
120 verify_oop(obj);
121 // test if object header is pointing to the displaced header, and if so, restore
122 // the displaced header in the object - if the object header is not pointing to
123 // the displaced header, get the object header instead
124 MacroAssembler::lock(); // must be immediately before cmpxchg!
125 cmpxchgptr(hdr, Address(obj, hdr_offset));
126 // if the object header was not pointing to the displaced header,
127 // we do unlocking via runtime call
128 jcc(Assembler::notEqual, slow_case);
129 // done
130 bind(done);
131
132 dec_held_monitor_count();
133 }
134
135
136 // Defines obj, preserves var_size_in_bytes
137 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
138 if (UseTLAB) {
139 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
140 } else {
141 jmp(slow_case);
142 }
143 }
144
145
146 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
147 assert_different_registers(obj, klass, len);
148 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
149 #ifdef _LP64
150 if (UseCompressedClassPointers) { // Take care not to kill klass
151 movptr(t1, klass);
152 encode_klass_not_null(t1, rscratch1);
153 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
154 } else
155 #endif
156 {
157 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
158 }
159
160 if (len->is_valid()) {
161 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
162 }
163 #ifdef _LP64
164 else if (UseCompressedClassPointers) {
165 xorptr(t1, t1);
166 store_klass_gap(obj, t1);
167 }
168 #endif
169 }
170
171
172 // preserves obj, destroys len_in_bytes
173 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
174 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
175 Label done;
176
177 // len_in_bytes is positive and ptr sized
178 subptr(len_in_bytes, hdr_size_in_bytes);
179 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
180 bind(done);
181 }
182
183
184 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
185 assert(obj == rax, "obj must be in rax, for cmpxchg");
186 assert_different_registers(obj, t1, t2); // XXX really?
187 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
188
189 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
190
191 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
192 }
193
194 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
195 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
196 "con_size_in_bytes is not multiple of alignment");
197 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
198
199 initialize_header(obj, klass, noreg, t1, t2);
200
201 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
202 // clear rest of allocated space
203 const Register t1_zero = t1;
204 const Register index = t2;
205 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
206 if (var_size_in_bytes != noreg) {
207 mov(index, var_size_in_bytes);
208 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
209 } else if (con_size_in_bytes <= threshold) {
210 // use explicit null stores
211 // code size = 2 + 3*n bytes (n = number of fields to clear)
212 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
213 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
214 movptr(Address(obj, i), t1_zero);
215 } else if (con_size_in_bytes > hdr_size_in_bytes) {
216 // use loop to null out the fields
217 // code size = 16 bytes for even n (n = number of fields to clear)
218 // initialize last object field first if odd number of fields
219 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
220 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
221 // initialize last object field if constant size is odd
222 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
223 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
224 // initialize remaining object fields: rdx is a multiple of 2
225 { Label loop;
226 bind(loop);
227 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
228 t1_zero);
229 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
230 t1_zero);)
231 decrement(index);
232 jcc(Assembler::notZero, loop);
233 }
234 }
235 }
236
237 if (CURRENT_ENV->dtrace_alloc_probes()) {
238 assert(obj == rax, "must be");
239 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
240 }
241
242 verify_oop(obj);
243 }
244
245 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
246 assert(obj == rax, "obj must be in rax, for cmpxchg");
247 assert_different_registers(obj, len, t1, t2, klass);
248
249 // determine alignment mask
250 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
251
252 // check for negative or excessive length
253 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
254 jcc(Assembler::above, slow_case);
255
256 const Register arr_size = t2; // okay to be the same
257 // align object end
258 movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
259 lea(arr_size, Address(arr_size, len, f));
260 andptr(arr_size, ~MinObjAlignmentInBytesMask);
261
262 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
263
264 initialize_header(obj, klass, len, t1, t2);
265
266 // clear rest of allocated space
267 const Register len_zero = len;
268 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
269
270 if (CURRENT_ENV->dtrace_alloc_probes()) {
271 assert(obj == rax, "must be");
272 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
273 }
274
275 verify_oop(obj);
276 }
277
278
279
280 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
281 verify_oop(receiver);
282 // explicit NULL check not needed since load from [klass_offset] causes a trap
283 // check against inline cache
284 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
285 int start_offset = offset();
286
287 if (UseCompressedClassPointers) {
288 load_klass(rscratch1, receiver, rscratch2);
289 cmpptr(rscratch1, iCache);
290 } else {
291 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
292 }
293 // if icache check fails, then jump to runtime routine
294 // Note: RECEIVER must still contain the receiver!
295 jump_cc(Assembler::notEqual,
296 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
297 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
298 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
299 }
300
301
302 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
303 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
304 // Make sure there is enough stack space for this method's activation.
305 // Note that we do this before doing an enter(). This matches the
306 // ordering of C2's stack overflow check / rsp decrement and allows
307 // the SharedRuntime stack overflow handling to be consistent
308 // between the two compilers.
309 generate_stack_overflow_check(bang_size_in_bytes);
310
311 push(rbp);
312 if (PreserveFramePointer) {
313 mov(rbp, rsp);
314 }
315 #if !defined(_LP64) && defined(COMPILER2)
316 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
317 // c2 leaves fpu stack dirty. Clean it on entry
318 empty_FPU_stack();
319 }
320 #endif // !_LP64 && COMPILER2
321 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
322
323 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
324 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
325 bs->nmethod_entry_barrier(this, NULL /* slow_path */, NULL /* continuation */);
326 }
327
328
329 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
330 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
331 pop(rbp);
332 }
333
334
335 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
336 if (breakAtEntry || VerifyFPU) {
337 // Verified Entry first instruction should be 5 bytes long for correct
338 // patching by patch_verified_entry().
339 //
340 // Breakpoint and VerifyFPU have one byte first instruction.
341 // Also first instruction will be one byte "push(rbp)" if stack banging
342 // code is not generated (see build_frame() above).
343 // For all these cases generate long instruction first.
344 fat_nop();
345 }
346 if (breakAtEntry) int3();
347 // build frame
348 IA32_ONLY( verify_FPU(0, "method_entry"); )
349 }
350
351 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
352 // rbp, + 0: link
353 // + 1: return address
354 // + 2: argument with offset 0
355 // + 3: argument with offset 1
356 // + 4: ...
357
358 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
359 }
360
361 #ifndef PRODUCT
362
363 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
364 if (!VerifyOops) return;
365 verify_oop_addr(Address(rsp, stack_offset));
366 }
367
368 void C1_MacroAssembler::verify_not_null_oop(Register r) {
369 if (!VerifyOops) return;
370 Label not_null;
371 testptr(r, r);
372 jcc(Assembler::notZero, not_null);
373 stop("non-null oop required");
374 bind(not_null);
375 verify_oop(r);
376 }
377
378 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
379 #ifdef ASSERT
380 if (inv_rax) movptr(rax, 0xDEAD);
381 if (inv_rbx) movptr(rbx, 0xDEAD);
382 if (inv_rcx) movptr(rcx, 0xDEAD);
383 if (inv_rdx) movptr(rdx, 0xDEAD);
384 if (inv_rsi) movptr(rsi, 0xDEAD);
385 if (inv_rdi) movptr(rdi, 0xDEAD);
386 #endif
387 }
388
389 #endif // ifndef PRODUCT